arya = {
    "role"        : "Software Development Engineer @ Salesforce",
    "location"    : "Bengaluru, Karnataka, India",
    "focus"       : ["LLM Engineering", "AI Agents", "MCP Servers", "Backend Systems"],
    "building"    : "Production-grade AI tools developers actually use every day",
    "certifications": "13× Anthropic Academy (Claude, MCP, Agents, Subagents, API)",
    "fun_fact"    : "I read the Anthropic docs for fun and then build things from them",
}

I'm a software engineer at Salesforce with a background in CV/ML from DTU, now deep in the world of LLMs, agents, and the Model Context Protocol. I spend evenings building side projects at the intersection of developer tooling and AI — things I wish existed when I was shipping features at work.

All completed March–April 2026 · Issued by Anthropic Education

Languages

AI / LLM Stack

Backend & Infrastructure

Tools & Platforms

Building production-grade AI tools, not toy demos.

A production RAG pipeline — ingest any PDF/doc collection, ask questions, get grounded answers with source citations. Includes hallucination detection, retrieval quality metrics, and a live monitoring dashboard. Deployed live on Railway.

Python · LangChain · ChromaDB · FastAPI · React · Railway

A GitHub Action + CLI that auto-reviews PRs using the Claude API — flags security issues, code smells, and suggests refactors with explanations. Published to npm so engineers can install it in 30 seconds.

TypeScript · Claude API · GitHub Actions · AST parsing · npm

An MCP server that connects Claude directly to Salesforce — query records, run SOQL, update objects, and trigger flows — all from within your AI assistant. Built with the official MCP SDK and published to npm. The first Salesforce MCP server on npm.

TypeScript · MCP SDK · Salesforce API · npm

Ingest a live data stream → clean → LLM enrichment (sentiment + entity extraction + summarization) → PostgreSQL → REST API. End-to-end system showing event-driven architecture, pipeline design, and production LLM integration.

Python · Kafka · PostgreSQL · FastAPI · Docker Compose · Claude API

A pytest-based framework for evaluating LLM outputs — factuality checks, latency benchmarks, hallucination detection, and regression testing. Integrates with DeepEval and RAGAS. Runs in CI via GitHub Actions.

Python · DeepEval · RAGAS · pytest · GitHub Actions

I write about building with LLMs, the Model Context Protocol, AI agents, and what I'm learning from Anthropic's stack. Find me on Medium →

If you're working on something interesting in the LLM/agent/MCP space, or hiring senior engineers for London-based teams, I'd love to chat.

📬 LinkedIn · Medium · X / Twitter